Fireworks AI in Talks to Raise at $15 Billion Valuation
Fireworks AI, which helps enterprises deploy and run AI models at scale, is in active talks to raise a new funding round at a $15 billion valuation, with Index Ventures set to co-lead.
Top Stories
All Sections
Finance
About
20 results for "AI models"
Fireworks AI, which helps enterprises deploy and run AI models at scale, is in active talks to raise a new funding round at a $15 billion valuation, with Index Ventures set to co-lead.
Google has made Gemini 3.5 Flash generally available at $1.50 per million input tokens, offering frontier-level reasoning at four times the speed of comparable models — putting direct price pressure on OpenAI's fast inference lineup.
Microsoft, xAI, and other major AI companies have agreed to provide U.S. government regulators early access to AI models before public release — the most substantive voluntary federal AI oversight commitment in the U.S. to date.
Google unveiled Gemini 3.5 Flash and the Gemini Spark agentic assistant at I/O 2026 — its most direct challenge to OpenAI and Anthropic in a single event.
Palo Alto Networks' May patch cycle found 26 CVEs using frontier AI models — versus the typical fewer than 5 per month. The window between vulnerability disclosure and weaponized exploit is shrinking fast.
OpenAI's new Daybreak initiative uses frontier AI models and Codex Security to proactively identify and validate patches for software vulnerabilities — entering the enterprise security market directly.
The Trump administration is considering requiring frontier AI labs to submit safety test results to the government — a reversal driven by national security concerns over Anthropic's Mythos model.
The Trump administration is weighing an executive order to create a formal government review process for AI models before release — triggered by Anthropic's Mythos, a model the company says is too dangerous to ship publicly.
Google DeepMind, Microsoft, and xAI signed agreements with NIST to provide pre-release frontier model access to US government evaluators. All five major US AI labs are now part of the program.
Anthropic has released Claude Opus 4.7, its latest flagship model, with significant improvements to multi-step reasoning and agentic workflow execution—the ability to plan and complete complex tasks autonomously across connected systems.
Cloud provider Nebius Group is acquiring 20-person inference optimization startup Eigen AI for $643 million, signaling that making AI models run cheaper is now worth as much as raw compute capacity.
A paper at ICLR 2026 finds that reinforcement-learning reasoning training — the technique behind o3 and Gemini Thinking — proportionally increases tool-call hallucinations as model performance improves.
Microsoft and OpenAI have amended their partnership, ending Microsoft's exclusive access to OpenAI models and freeing OpenAI to deploy on any cloud — including AWS, Google Cloud, and Oracle.
The Trump administration announced plans to crack down on Chinese companies exploiting U.S. AI models — an escalation that lands on the same day DeepSeek released its most powerful model yet.
OpenAI released GPT-5.5 on April 23—a model built to act, not just answer, capable of autonomously chaining tools across coding, computer use, and deep research without waiting for user prompts.
DeepSeek released V4 Flash and V4 Pro on April 24 with top-tier coding benchmarks and a 1-million-token context window—open-source, again, at a moment when Silicon Valley assumed the gap was widening.
Frontier AI models now solve real software engineering tasks with near-perfect accuracy — but the same report finds leading AI systems are disclosing less about how they work than ever before.
NVIDIA's open Ising model family delivers error-correction decoding that is 2.5x faster and 3x more accurate than traditional methods, removing the calibration and decoding bottlenecks blocking practical quantum computing.
NVIDIA released Ising on World Quantum Day — the first open-source family of AI models built specifically for quantum computing, including a 35B-parameter vision-language model for processor calibration and a decoder that runs 2.5x faster than classical error correction methods.
Meta released Muse Spark, claiming it matches Llama 4 capability at one-tenth the training compute — a potential inflection point in AI efficiency economics.